what is taste

An AI gives you twenty drafts of the same email. None are wrong. You pick one. The picking is what taste is.

a working definition

Taste is the skill of picking what's good when no rule tells you how.

When a rule is enough, you don't need taste, you need execution. A recipe says 250g of flour and you weigh it out. The linter catches your syntax error. The interview rubric scores eight out of ten.

You use taste where the rule runs out:

Picture the space of answers you could give. Most of it is bad. Inside the bad sits a smaller region of good answers, and inside that, the answers a rule can take you to.

Craft is being fast and good inside the rule region. You learn the rules well enough to vary them, and you move from one good rule-reachable answer to a better one. That accounts for most of skill development.

Taste is what gets you to the good answers outside the rule region. Some of those, the rules can't reach.

same machinery, different output

• Cooking. The recipe gives you ratios. The chef tastes the sauce, adds a half-pinch more salt, decides it's done. The recipe didn't say when.
• Code review. The linter passed. Tests passed. The reviewer writes "this abstraction is wrong" and can't put it in a checklist.
• Hiring. Both candidates passed the structured interview. You pick one. The rubric didn't decide for you.
• Editing video. Three takes of the same line, all technically clean. The editor keeps take two and can't say what makes it the one.
• Designing a page. The grid is mathematically right. The page feels dead. The designer moves the headline down 12px and the page wakes up.
• Writing. Grammar fine. Spelling fine. The paragraph sits there. You kill three sentences and it stands up.

but it doesn't transfer

The same machinery does not produce the same taste. A venture investor with strong taste for companies has no special insight into which products those companies should build. They notice different things: business durability, founder traits, distribution moats. A product designer notices a different set. An engineer who builds well has no special claim on what to build. A designer who picks the cleanest layout has no special claim on hiring. Scramble the board into illegal positions and chess masters lose their memory advantage (Chase & Simon, 1973): what they remember is meaningful chess, not visual patterns. Deliberate practice builds skill on the task you practice, not on adjacent ones (Ericsson et al., 1993).

Transfer happens in residues. A designer can judge a poster, a book cover, a film frame, adjacent visual problems where the schemas overlap. The same designer picking which company to invest in: no carry. The closer the perceptual demands of two domains, the more taste transfers between them. The further apart they are, the less. The mechanisms underneath are domain-specific: perceptual learning tunes you to the inputs you practiced on (Goldstone, 1998).

The same three pieces show up in every room: perception, standards, judgement. You calibrate them in each room on that room's examples, against that room's underspecified rubric, with that room's "good." The machinery generalises. The taste you build with it transfers to adjacent rooms, not further.

Most of the literature is visual. Researchers derived the framework from paintings, faces, photographs, landscapes. The predictions are sharpest there. For sound, for ideas, for problems, for businesses, the three pieces hold. The evidence thins.

the three traditions

Three traditions have argued about what taste is. Kant first, Bourdieu second, the psychologists third.

Kant: judgment with universal pretension

Kant's Critique of the Power of Judgment (Kant, 1790) treats taste as a faculty that judges the beautiful. The judgment is subjective: you feel it before you reason about it. It is disinterested: you don't want to own the thing. And it claims universal validity: when you call something beautiful, you expect everyone else to call it that too. Kant spent the third Critique defending that combination.

Preference and judgment work differently. A preference doesn't claim universal agreement. Say "I like vanilla" and nobody else has to agree. A taste judgment does claim it. Say "this scene needs strings, not piano" and you expect everyone paying attention to agree. The expectation of agreement is what makes it a judgment.

Bourdieu: taste as a social marker

Pierre Bourdieu's Distinction (Bourdieu, 1984) treats taste as a social marker. Your taste signals your class. The mechanism is habitus: a set of dispositions you absorbed in childhood that shape what you perceive and what you judge before you start to think about it.

People from different class backgrounds like reliably different things. You can't see where your taste came from. What feels like personal preference is class calibration.

Cultural capital comes in three forms (Bourdieu, 1986). Embodied: how you carry yourself, what feels obvious, what feels embarrassing. Objectified: the books, the instruments, the art on your wall. Institutionalised: the degree on your wall. Embodied capital takes the longest because no one can give it to you. You absorb it.

The original Bourdieu has aged unevenly. He framed high-status taste as exclusive: snobbery as the signal. Peterson & Kern (1996) showed the signal flipped. Liking a bit of everything, high and low, now does what snobbery used to do. The mechanism still holds; the operating rule changed.

Empirical aesthetics: the cognitive machinery

Psychology has spent thirty years measuring what taste does. The work that matters here splits into three claims.

The verdict and the feeling are separate. Leder et al. (2004) drew the line; Leder & Nadal (2014) tightened it a decade later. The verdict is what you say about the work ("this is excellent"). The feeling is what the work does to you (pleasure, being moved, indifference). The two can come apart. A film can be badly made and still a great time. A masterpiece can leave you cold. A single rating misses the gap.

The gap is the off-diagonal of a 2×2: the cells where the verdict and the feeling disagree.

Liking tracks ease of processing. Reber et al. (2004) showed that when your perceptual system handles a stimulus without effort, that ease registers as pleasure. You feel the thing is good. The pleasure was in the processing. The same mechanism explains the mere-exposure effect: repeated exposure to almost anything raises your liking for it, one of the most replicated findings in psychology (Zajonc, 1968; Bornstein, 1989; Montoya et al., 2017).

The effect runs as an inverted-U over exposures. A few exposures do nothing. Sustained exposure builds liking. Past the peak, boredom takes over.

Apply that to every answer at once. The rule-reachable peak saturates first because most people can apply the rules of the moment, so that work shows up everywhere. The peak outside the rules holds longer, until it gets imitated enough to be absorbed. The rule region drifts to cover what taste used to be. A new peak forms further out.

Taste is a moving target. Yesterday's leap is today's rule.

Some preferences travel across cultures. Reviews from Brown (2018) and Palmer, Schloss & Sammartino (2013, Annual Review of Psychology) catalogue a small cluster that replicates everywhere it's been tested. Symmetric faces (Rhodes, 2006, meta-analysis). Averaged composites (Langlois et al., 2000, meta-analysis). Curves over angles at 84-millisecond presentations (Bar & Neta, 2006; the mechanism is amygdala threat-detection, separate from fluency). Mid-range fractal complexity at D ≈ 1.3–1.5 (Spehar et al., 2003). Park-like landscapes over forests or deserts (Kaplan & Kaplan, 1989).

Others don't replicate. McDermott et al. (2016) tested the Tsimane' of the Bolivian Amazon on the consonance-over-dissonance preference. The Tsimane' rate consonant and dissonant chords as equally pleasant. The golden ratio fails modern replications. Many colour preferences are culture-shaped.

The picture you end up with is a small set of innate preferences plus cultural and personal layers on top. Your aesthetic response has a biological floor and a learned ceiling.

What each tradition gets right

Kant explains what taste claims to be doing: judging, while expecting others to agree. Bourdieu explains what taste does socially: marking position. Psychology explains the machinery: the schemas in your head, what feels easy, what's cross-cultural, what you learned on top. None of them is the whole answer.

what's underneath

To develop taste, three pieces matter.

The three pieces map onto the empirical work. Perception is what expertise builds: Chase & Simon (1973) showed it in chess masters, and the same chunking happens in wine tasters, code reviewers, editors. Standards are schemas: internal models of "good" structured enough to apply to new cases (Bartlett, 1932). Judgment is what Leder calls the verdict: the moment you pick, with the criteria not fully settled (Leder et al., 2004).

You need all three. The sommelier who names twenty wine faults but freezes on which one to recommend has perception, no judgment. The senior designer with strong opinions on every page who can't see when their own work is overdesigned has standards, no perception. The PM who decides fast but always picks safer has judgment, no standards.

Strong taste develops all three at once. That's rare, which is why the people who have it are people you can name, not roles.

why it matters now

For most of history, the bottleneck for making something was upstream of the choosing:

In each era the scarce skill was producing: grinding the pigment, operating the press, getting on the radio, catching the algorithm. While generation was hard, taste sat downstream as a luxury.

Generation is no longer the bottleneck. Ask a model for twenty headlines or twenty mood boards. The returns come in three seconds. None are wrong. You still ship one.

AI will generate everything. Choosing what to keep is your part.